Medical automation research method and device
By integrating academic retrieval and natural language processing technologies, the system automatically acquires literature and filters research methods, automatically generates experimental scripts and LaTeX documents, solving the problem of low efficiency in existing medical research processes and achieving efficient and professional research automation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING HUIMEI CLOUD TECHNOLOGY CO LTD
- Filing Date
- 2026-04-29
- Publication Date
- 2026-06-05
AI Technical Summary
Current medical research processes rely on manual operations, which are inefficient and prone to overlooking cutting-edge technologies. Literature acquisition and processing are not automated, experiments are inefficient, and generated documents are difficult to reproduce and lack professionalism.
By integrating academic retrieval tool interfaces to automatically acquire literature, combining natural language processing and multi-dimensional reliability assessment models to screen research methods, automatically generating experimental scripts and calling computing resources to execute experiments, establishing alignment between research methods and code implementation, and training a document generation model to generate academic documents in LaTeX format.
It significantly improves the efficiency of literature processing, selects high-quality research plans, reduces invalid experiments, ensures the professionalism and reproducibility of generated documents, and improves research efficiency and document quality.
Smart Images
Figure CN122154632A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical information technology, and in particular to a medical automation research method and apparatus. Background Technology
[0002] In the current medical research process, researchers usually need to manually search for literature, screen methods, design experiments and write papers. The whole process is highly dependent on human experience, which is inefficient and prone to overlooking cutting-edge technologies.
[0003] With the rapid development of artificial intelligence technology, some studies have attempted to apply large language models to scientific research assistance, such as training scientific research agents to generate paper outlines, experimental designs, and automatically generate LaTeX format papers.
[0004] However, existing technologies still require manual screening of solutions during the literature input stage, which cannot achieve automated acquisition and standardized processing, thus limiting the scale and intelligence of the scientific research process. Moreover, existing technologies often directly copy methods for experiments after obtaining literature, without quantitatively evaluating the innovativeness, reproducibility, and significance of the results. This results in a large number of ineffective experiments consuming resources and makes it difficult to quickly identify high-value research paths. In addition, directly calling a general large model to generate LaTeX documents can easily lead to inaccurate descriptions of professional content (such as formulas, algorithm logic, and experimental parameters), and the generated documents are difficult to align with the code implementation, making them unsuitable for direct experimental reproduction or model training. Summary of the Invention
[0005] Therefore, it is necessary to provide a medical automation research method and device to address the aforementioned technical problems. By integrating academic search tool interfaces or using agent skills methods for browsing web pages, it can automatically obtain literature related to the target task and extract research methods and parameters by combining natural language processing technology, thereby significantly improving the efficiency of literature processing and reducing manual intervention.
[0006] This invention provides a medical automation research method, the method comprising: Automatically retrieve academic papers related to the target medical research task and construct an initial literature database; Research methods and their experimental parameters are extracted from the papers in the initial literature database, and the research methods are screened based on a preset reliability assessment model to obtain a candidate solution database. Based on the research methods and experimental parameters in the candidate scheme library, an experimental script is automatically generated and computing resources are called to execute the experiment and obtain the experimental results. Based on the research methods in the candidate solution library, obtain the corresponding code implementations, establish the alignment relationship between the research methods and the code implementations, and use the alignment data to train a document generation model. When the experimental results meet the preset conditions, the research methods and experimental results are integrated using the document generation model to generate an academic document in LaTeX format.
[0007] In one embodiment, the automatic retrieval of academic papers related to the target medical research task and the construction of an initial literature database include: Set search criteria, which include research field, time range, and keywords; By calling the application programming interface of academic search tools or using the agent skills method of browsing pages, relevant papers can be automatically retrieved according to the search criteria, and paper metadata can be obtained. Download the full text of the paper based on its metadata, extract the text content of the paper, and build an initial literature database.
[0008] In one embodiment, the step of extracting research methods and their experimental parameters from papers in the initial literature database, and screening the research methods based on a preset reliability assessment model to obtain a candidate solution library, includes: Natural language processing technology is used to parse the text of papers in the initial literature database, and to identify and extract the research methods and experimental parameters described therein. The extracted research methods are evaluated in multiple dimensions based on a pre-defined reliability assessment model. The evaluation dimensions include methodological innovation, experimental reproducibility, and result significance. Based on the evaluation results, the research methods are ranked, and methods that meet the reliability threshold and their corresponding experimental parameters are selected to form a candidate scheme library.
[0009] In one embodiment, the step of automatically generating experimental scripts and calling computing resources to execute experiments and obtain experimental results based on research methods and experimental parameters in the candidate solution library includes: The experimental parameters of each research method in the candidate scheme library are analyzed, and the experimental scripts in the corresponding programming languages are generated. The experimental script is executed by calling the computing resource cluster, and training logs and intermediate results are collected in real time during the execution process; After the experiment is completed, the preset evaluation index values are extracted from the output as the experimental results.
[0010] In one embodiment, obtaining the corresponding code implementation based on the research methods in the candidate solution library includes: Collect academic papers in the field of artificial intelligence and extract the addresses of publicly available code repositories by parsing the text of the papers. Based on the research methods in the candidate solution library, match the corresponding code repository address, and clone or download the code repository; Parse the directory structure and dependencies of the code repository to generate a code structure description.
[0011] In one embodiment, establishing the alignment between research methods and code implementation, and training a document generation model using aligned data, includes: Establish a correspondence between research methods and code implementations, and combine the research method description text and code snippets into training samples; The pre-trained language model is fine-tuned using the training samples, with the training objective being to generate corresponding LaTeX documents based on the research methodology description.
[0012] In one embodiment, when the experimental results meet preset conditions, the document generation model is used to integrate the research methods and experimental results to generate an academic document in LaTeX format, including: Determine whether the experimental results exceed the preset performance improvement threshold; If so, the trained document generation model is invoked, and the research method description and experimental results are input to generate a first draft of an academic document in LaTeX format. Adjust the format and verify the content of the first draft, and output the final academic document.
[0013] The present invention also provides a medical automated research device, applied to the medical automated research method described in any of the above embodiments, the device comprising: The automatic retrieval module is used to automatically retrieve academic papers related to the target medical research task and build an initial literature database. The screening module is used to extract research methods and their experimental parameters from the papers in the initial literature database, and to screen the research methods based on a preset reliability assessment model to obtain a candidate scheme library. The execution module is used to automatically generate experimental scripts and call computing resources to execute experiments based on the research methods and experimental parameters in the candidate scheme library, and obtain experimental results; The training module is used to obtain the corresponding code implementation based on the research methods in the candidate solution library, establish the alignment relationship between the research methods and the code implementation, and use the alignment data to train the document generation model. An automatic generation module is used to integrate the research methods and experimental results into an academic document in LaTeX format when the experimental results meet preset conditions, using the document generation model.
[0014] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the medical automation research method as described above.
[0015] The present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, implements the medical automation research method as described above.
[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the medical automation research method as described above.
[0017] The aforementioned automated medical research methods and devices, by integrating academic retrieval tool interfaces, automatically acquire literature related to the target task and extract research methods and parameters using natural language processing technology, significantly improving literature processing efficiency and reducing manual intervention. By introducing a multi-dimensional reliability assessment model, the innovativeness, reproducibility, and significance of research methods are quantitatively evaluated, selecting high-quality solutions, avoiding invalid experiments, and improving research efficiency. By linking to open-source platforms such as GitHub, the corresponding code implementations are automatically acquired, and an alignment relationship between research methods and code snippets is constructed. A dedicated LaTeX document generation model is trained to ensure a high degree of matching in terms of professionalism and format. The document generation model trained based on the aligned data can automatically generate LaTeX documents that conform to academic standards according to the research methods and experimental results, reducing human writing bias and improving document readability and reproducibility. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0019] Figure 1 A flowchart of a medical automation research method provided by the present invention; Figure 2 A modular diagram of a medical automated research device provided by the present invention; Figure 3 An internal structural diagram of the computer device provided by the present invention. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0021] The following is combined Figures 1 to 3 This invention describes a medical automation research method and apparatus.
[0022] In one embodiment, a medical automation research method includes the following steps: Step S100: Automatically retrieve academic papers related to the target medical research task and construct an initial literature database.
[0023] Step S200: Extract research methods and their experimental parameters from the papers in the initial literature database, and screen the research methods based on a preset reliability assessment model to obtain a candidate solution database.
[0024] Step S300: Based on the research methods and experimental parameters in the candidate scheme library, automatically generate experimental scripts and call computing resources to execute experiments and obtain experimental results.
[0025] Step S400: Based on the research methods in the candidate solution library, obtain the corresponding code implementations, establish the alignment relationship between the research methods and the code implementations, and use the alignment data to train a document generation model.
[0026] Step S500: When the experimental results meet the preset conditions, the research methods and experimental results are integrated using the document generation model to generate an academic document in LaTeX format.
[0027] The aforementioned automated medical research methods, by integrating academic retrieval tool interfaces, automatically acquire literature relevant to the target task and extract research methods and parameters using natural language processing technology, significantly improving literature processing efficiency and reducing manual intervention. By introducing a multi-dimensional reliability assessment model, the innovativeness, reproducibility, and significance of research methods are quantitatively evaluated, selecting high-quality solutions, avoiding invalid experiments, and improving research efficiency. By linking to open-source platforms such as GitHub, the methods automatically acquire corresponding code implementations and establish alignment relationships between research methods and code snippets, training a dedicated LaTeX document generation model to ensure a high degree of matching in terms of professionalism and format. The document generation model, trained based on aligned data, can automatically generate academically compliant LaTeX documents according to research methods and experimental results, reducing human writing bias and improving document readability and reproducibility.
[0028] In one embodiment, the automatic retrieval of academic papers related to the target medical research task and the construction of an initial literature database includes the following steps: Step S110: Set search criteria, which include research field, time range and keywords.
[0029] Specifically, the target medical research task is set as "Research on a Deep Learning-Based CT Image-Assisted Diagnostic Model for Lung Cancer". To achieve automatic literature retrieval, it is first necessary to set search criteria.
[0030] Search criteria include research area, time range, and keywords, among which: The research area is limited to medical image analysis and deep learning, specifically for lung cancer diagnosis; Keywords may include "lung cancer", "CT imaging", "deep learning", "convolutional neural network", etc. The time frame is set to the last five years, such as 2019 to 2024, to ensure that the obtained documents are timely.
[0031] These search criteria can be entered through system configuration files or the user interface, serving as the basis for subsequent automated searches.
[0032] Step S120: Call the application programming interface of the academic search tool or use the agentskills method of the browsing page to automatically retrieve relevant papers and obtain paper metadata according to the search conditions.
[0033] Specifically, based on the set search criteria, the system calls the application programming interface (API) of academic search tools to perform automatic searches. Available academic search tools include the arXiv API and the PubMed Central API.
[0034] Pass search criteria (such as keywords and time range) as parameters to the API to retrieve metadata for matching papers. Metadata includes information such as paper title, authors, abstract, publication date, DOI, and PDF download link.
[0035] Next, the returned metadata is deduplicated and temporarily stored to prepare for subsequent full-text downloads. This process can be executed periodically to ensure continuous updates of the literature.
[0036] Step S130: Download the full text of the paper based on the paper's metadata, extract the text content of the paper, and construct an initial literature database.
[0037] Specifically, the full text of the paper is automatically downloaded based on the PDF link in the obtained paper metadata. For papers requiring access permissions, processing can be performed according to preset institutional access credentials.
[0038] After downloading, use a PDF parsing tool (such as GROBID or PyMuPDF) to extract the text content from the paper, focusing on extracting the abstract, methods, experiments, conclusions, and other relevant sections.
[0039] The extracted text content, along with metadata, is stored in the database to form a structured initial literature repository. For example, the system can download and parse 50 relevant papers to build an initial literature repository containing the full text of the papers and their metadata.
[0040] In one embodiment, the step of extracting research methods and their experimental parameters from papers in the initial literature database, and screening the research methods based on a preset reliability assessment model to obtain a candidate solution library, includes the following steps: Step S210: Use natural language processing technology to parse the text of the papers in the initial literature database, identify and extract the research methods and experimental parameters described therein.
[0041] Specifically, the text of papers in the initial literature database is parsed, and natural language processing technology is used to identify and extract research methods and their experimental parameters.
[0042] For example, a BERT-based named entity recognition and relation extraction model can be used to perform semantic analysis on the "Methods" section of the paper to extract model names (such as "3D ResNet" and "EfficientNet-B7") and experimental parameters (such as batch size, learning rate, optimizer, dataset name, etc.).
[0043] The extracted information is stored in a structured form, forming a record of the corresponding research methods and parameters.
[0044] For example, information such as "Method name: Multi-scale 3D CNN, Parameters: Input size 128×128×64, Loss function: Cross-entropy, Number of training rounds: 100" can be extracted from a certain paper.
[0045] Step S220: The extracted research methods are evaluated in multiple dimensions based on a preset reliability assessment model. The evaluation dimensions include method innovation, experimental reproducibility, and result significance.
[0046] Specifically, the pre-defined reliability assessment model is a multi-dimensional quantitative assessment model for medical research methods, constructed using the analytic hierarchy process (AHP) plus linear weighted scoring, for example: Based on the consensus of experts in the field of medical research, three core evaluation dimensions were identified: methodological innovation, experimental reproducibility, and result significance. Each dimension was further broken down into quantifiable secondary indicators to form a standardized evaluation system.
[0047] The input consists of structured data extracted from the paper, including research method text, experimental parameters, code publication status, dataset publication status, performance metrics, citation data, and method differences.
[0048] The model calculation process includes: standardizing each secondary indicator to a score of 0-1; weighting the secondary indicators according to preset weights to obtain a single-dimensional score; summing the weighted scores of the three dimensions to obtain a comprehensive reliability score; and outputting the score results for subsequent sorting and filtering.
[0049] Output single-dimensional scores and comprehensive reliability scores (0-1 points). Research methods with scores ≥ preset reliability thresholds are included in the candidate solution library.
[0050] The model was trained and optimized using over 1000 annotated papers in the medical research field. The weights were calibrated using a combination of domain expert scoring and entropy weighting, making it reproducible and universal.
[0051] The system invokes a pre-defined reliability assessment model to perform a multi-dimensional evaluation of each extracted research method.
[0052] The evaluation dimensions include: methodological innovation, experimental reproducibility, and result significance, among which: Methodological innovation can be quantified by analyzing paper citation relationships and the degree of difference from existing methods; Experiment reproducibility is judged based on indicators such as whether the code is publicly available, whether the dataset is publicly available, and whether the parameter description is complete. The significance of the results is evaluated based on whether the performance metrics (such as AUC and accuracy) reported in the paper are superior to the baseline model.
[0053] Each dimension is assigned a corresponding weight, and a comprehensive reliability score is calculated. For example, a paper that simultaneously publishes its code on GitHub and achieves an AUC of 0.97 on a public dataset will have a higher reliability score.
[0054] Specific examples are as follows:
[0055] Methodological innovation (0-1 points): No existing method = 1 point; Significant difference from existing methods = 0.8 points; Minor improvement = 0.5 points; No innovation = 0 points.
[0056] Experiment reproducibility (0-1 points): Public code + public dataset + complete parameters = 1 point; Public code + complete parameters = 0.7 points; Complete parameters only = 0.4 points; Missing parameters = 0 points.
[0057] Significance of results (0-1 points): Performance exceeding baseline by more than 10% = 1 point; exceeding baseline by 5%-10% = 0.7 points; exceeding baseline by 0-5% = 0.4 points; below baseline = 0 points.
[0058] Overall reliability score = Innovation score × 30% + Reproducibility score × 40% + Result significance score × 30%.
[0059] The calculation example is as follows: A research method for medical imaging diagnosis: Methodological innovation: Minor improvement → 0.5 points Experiment reproducibility: Public code + complete parameters → 0.7 points Significance of results: Performance exceeded baseline by 6% → 0.7 points Overall score = 0.5 × 30% + 0.7 × 40% + 0.7 × 30% = 0.64 points The preset threshold is 0.6 points. This method meets the requirements and is included in the candidate solution library.
[0060] Step S230: Sort the research methods according to the evaluation results, select the methods that meet the reliability threshold and their corresponding experimental parameters, and form a candidate scheme library.
[0061] Specifically, based on the reliability assessment results, all research methods are ranked, and methods with scores higher than a preset threshold (e.g., 0.75) and their corresponding experimental parameters are selected to form a candidate scheme library.
[0062] Each solution in the candidate pool includes a method description, experimental parameters, paper source, and code address (if applicable). For example, 10 highly reliable methods can be selected from 50 papers to form the candidate pool for subsequent experiments. This selection mechanism effectively avoids the experimental waste associated with low-value solutions.
[0063] In one embodiment, the step of automatically generating an experimental script and calling computing resources to execute the experiment and obtain experimental results based on the research methods and experimental parameters in the candidate scheme library includes the following steps: Step S310: parse the experimental parameters of each research method in the candidate scheme library and generate the experimental script in the corresponding programming language.
[0064] Specifically, the experimental parameters of each method in the candidate solution library are analyzed, and combined with the preset code template, the experimental scripts in the corresponding programming languages are automatically generated.
[0065] For example, if the candidate method is implemented based on PyTorch, a Python script containing modules such as data loading, model definition, training loop, and validation evaluation will be generated based on the parameters.
[0066] Hyperparameters in the script (such as learning rate and batch size) are automatically filled in based on the extracted parameters, ensuring the script's executability. Each candidate method corresponds to an independent experimental script.
[0067] Step S320: Call the computing resource cluster to execute the experimental script, and collect training logs and intermediate results in real time during the execution process.
[0068] Specifically, the generated experimental scripts are submitted to the computing resource cluster for execution via the task scheduling module. The computing resources can be internal GPU servers or cloud GPU instances (such as Alibaba Cloud gn7 series).
[0069] Resources are allocated to each experiment, its running status is monitored, and training logs, intermediate results (such as loss values and validation accuracy in each round), model weights, and other data are collected in real time during execution. The collected data is stored in a designated directory for subsequent analysis and result extraction.
[0070] Step S330: After the experiment is completed, extract the preset evaluation index values from the output as the experimental results.
[0071] Specifically, after the experiment is completed, the pre-defined evaluation metrics are extracted from the output file (such as the training log) as the experimental results. The evaluation metrics are predefined according to the task type, such as accuracy, AUC, and F1-score in classification tasks.
[0072] Then, the log files are parsed to extract the best or final metric, which is then associated with and stored as the corresponding candidate method. For example, an experiment might extract an AUC of 0.973 with an accuracy of 91.2%, which is then used as the experimental result for that method.
[0073] In one embodiment, obtaining the corresponding code implementation based on the research methods in the candidate solution library includes the following steps: Step S410: Collect academic papers in the field of artificial intelligence and extract the addresses of publicly available code repositories by parsing the paper text.
[0074] Specifically, to train the document generation model, it is necessary to construct an aligned dataset of research methods and code implementations.
[0075] First, iterate through the paper lists of major conferences (such as NeurIPS, CVPR, MICCAI) and journals (such as TMI, MIA) in the field of artificial intelligence, download the PDFs and parse the text, and extract the publicly available GitHub code repository addresses in the papers using regular expressions or trained models.
[0076] The extracted URLs, along with the paper title, DOI, and other information, are stored in a paper-to-code mapping library. This process can be performed periodically to expand the dataset size.
[0077] Step S420: Match the corresponding code repository address according to the research methods in the candidate solution library, and clone or download the code repository.
[0078] Specifically, based on the research methods in the candidate solution library, the system searches for matching code repository addresses in the paper-code mapping library using information such as paper title and DOI. If a match is found, the `git clone` command is automatically executed to download the complete code repository to the local machine.
[0079] Step S430: Parse the directory structure and dependencies of the code repository to generate a code structure description.
[0080] Specifically, the downloaded code repository is parsed, the directory structure is traversed, and key files (such as the main training script, model definition files, and dependency files) are identified. Simultaneously, dependency files (such as requirements.txt) are parsed to obtain runtime environment information.
[0081] Then, a code structure description is generated, including a list of files, main functional modules, dependencies, etc., to provide context for subsequent alignment training. For example, the parsed description is: "The main training script is train.py, the model definition is located in model.py, and it depends on PyTorch 1.9, numpy, scikit-learn, etc."
[0082] In one embodiment, establishing the alignment between research methods and code implementation, and training a document generation model using aligned data, includes the following steps: Step S440: Construct the correspondence between research methods and code implementations, and combine the research method description text and code snippets into training samples.
[0083] Specifically, the research method description text is combined with the corresponding code snippets to form training samples. The research method description text can be extracted from the paper (e.g., "Propose a multi-scale 3D convolutional neural network for lung cancer feature extraction"), while the code snippets are extracted from the cloned repository to implement the method (e.g., the network definition code in model.py).
[0084] In addition, the LaTeX description fragments of the corresponding methods are extracted from the original paper as the target output; the research method description text, code implementation fragments, and LaTeX description fragments are aligned one-to-one to form a set of aligned samples from research method description, code to LaTeX; multiple sets of aligned samples form the model training dataset.
[0085] Step S450: Supervised fine-tuning of the pre-trained language model is performed using the training samples. The training objective is to generate the corresponding LaTeX document according to the research method description.
[0086] Specifically, the pre-trained language models (such as CodeT5, LLaMA, and ChatGLM) are supervised fine-tuned using the constructed alignment dataset.
[0087] Specific examples are as follows: Pre-trained language models such as CodeT5, LLaMA-2, and ChatGLM3, adapted for code and text generation, were selected.
[0088] Low-rank adaptation (LoRA) fine-tuning is employed (balancing performance and efficiency), with optional full-parameter fine-tuning (for high-precision scenarios).
[0089] Fine-tuning the core parameters includes: LoRA rank: r=8; LoRAα: 16; Dropout: 0.1 Learning rate: 2e-5; Batch size: 8; Number of training epochs: 10 Optimizer: AdamW; Loss function: Cross-entropy loss (text generation, sequence matching loss) The fine-tuning process includes: Input: "Description of research methods, code snippet"; Output: Standard LaTeX format text of the paper (methods section, formulas, parameters, results tables); The model learns medical research terminology, LaTeX syntax, and formula formatting rules to achieve a mapping from input to target output.
[0090] The fine-tuning goal is to input a natural language description of the research method and its corresponding code, and for the model to automatically generate a LaTeX document fragment conforming to the MICCAI / IEEE journal specifications, including formulas, experimental parameters, and results descriptions.
[0091] In one embodiment, when the experimental results meet preset conditions, the document generation model is used to integrate the research methods and experimental results to generate an academic document in LaTeX format, including the following steps: Step S510: Determine whether the experimental results exceed the preset performance improvement threshold.
[0092] Specifically, for methods in the candidate solution library that have already undergone experiments, it is determined whether their experimental results meet preset conditions. These preset conditions can be set as the improvement in performance metrics relative to the baseline model, such as an AUC improvement exceeding 5%.
[0093] If the experimental results of a method meet this condition, it is considered a valuable research result and enters the document generation process. For example, if a method's AUC improves from a baseline of 0.92 to 0.973, an improvement of approximately 5.8%, it meets the condition.
[0094] Step S520: If so, call the trained document generation model, input the research method description and experimental results, and generate a first draft of the academic document in LaTeX format.
[0095] Specifically, for methods that meet the criteria, the pre-trained document generation model is invoked, and the research method description and experimental results are input. The model then automatically generates a draft academic document in LaTeX format.
[0096] The initial draft includes sections such as title, authors (which can be preset), abstract, introduction, methods, experiments, results, conclusions, and references, all written in LaTeX syntax. For example, the methods section automatically inserts formulas (\begin{equation}...\end{equation}), and experimental results are presented in tabular form.
[0097] Step S530: Adjust the format and verify the content of the first draft, and output the final academic document.
[0098] Specifically, the generated LaTeX draft undergoes automated format checks and adjustments. These checks include formula completeness, citation validity, and correct figure / table labels, ensuring the format conforms to the template requirements of the target journal or conference (such as MICCAI or IEEE templates).
[0099] If a formatting issue is detected, the system will attempt to automatically repair it or prompt for manual intervention. After adjustments are complete, the final .tex file will be output, and a LaTeX compiler (such as pdflatex) can be used to generate a PDF document for researchers to review or for direct submission.
[0100] The medical automated research device provided by the present invention is described below. The medical automated research device described below and the medical automated research method described above can be referred to in correspondence.
[0101] In one embodiment, a medical automated research device is applied to the medical automated research method described in any of the above embodiments, including an automatic retrieval module, a screening module, an execution module, a training module, and an automatic generation module.
[0102] The automatic retrieval module is used to automatically retrieve academic papers related to the target medical research task and build an initial literature database.
[0103] The screening module is used to extract research methods and their experimental parameters from the papers in the initial literature database, and to screen the research methods based on a preset reliability assessment model to obtain a candidate scheme library.
[0104] The execution module is used to automatically generate experimental scripts and call computing resources to execute experiments based on the research methods and experimental parameters in the candidate scheme library, and obtain experimental results.
[0105] The training module is used to obtain the corresponding code implementation based on the research methods in the candidate solution library, establish the alignment relationship between the research methods and the code implementation, and use the alignment data to train the document generation model.
[0106] The automatic generation module is used to integrate the research methods and experimental results into an academic document in LaTeX format when the experimental results meet the preset conditions, using the document generation model.
[0107] Figure 3 This example illustrates a schematic diagram of the physical structure of an electronic device, which can be a smart terminal. Its internal structure diagram can be as follows: Figure 3 As shown, this electronic device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements automated medical research methods.
[0108] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the electronic device to which the present invention is applied. A specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0109] On the other hand, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, realizes a medical automation research method.
[0110] On another front, a computer program product or computer program is provided, comprising computer instructions stored in a computer storage medium. A processor of an electronic device reads the computer instructions from the computer storage medium, and when the processor executes the computer instructions, it implements a medical automation research method.
[0111] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.
[0112] By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0113] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0114] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.
Claims
1. A medical automation research method, characterized in that, The method includes: Automatically retrieve academic papers related to the target medical research task and construct an initial literature database; Research methods and their experimental parameters are extracted from the papers in the initial literature database, and the research methods are screened based on a preset reliability assessment model to obtain a candidate solution database. Based on the research methods and experimental parameters in the candidate scheme library, an experimental script is automatically generated and computing resources are called to execute the experiment and obtain the experimental results. Based on the research methods in the candidate solution library, obtain the corresponding code implementations, establish the alignment relationship between the research methods and the code implementations, and use the alignment data to train a document generation model. When the experimental results meet the preset conditions, the research methods and experimental results are integrated using the document generation model to generate an academic document in LaTeX format.
2. The medical automation research method according to claim 1, characterized in that, The automatic retrieval of academic papers related to the target medical research task constructs an initial literature database, including: Set search criteria, which include research field, time range, and keywords; By calling the application programming interface of academic search tools or using the agent skills method of browsing pages, relevant papers can be automatically retrieved according to the search criteria, and paper metadata can be obtained. Download the full text of the paper based on its metadata, extract the text content of the paper, and build an initial literature database.
3. The medical automation research method according to claim 2, characterized in that, The research methods and their experimental parameters are extracted from papers in the initial literature database, and the research methods are screened based on a preset reliability assessment model to obtain a candidate solution library, including: Natural language processing technology is used to parse the text of papers in the initial literature database, and to identify and extract the research methods and experimental parameters described therein. The extracted research methods are evaluated in multiple dimensions based on a pre-defined reliability assessment model. The evaluation dimensions include methodological innovation, experimental reproducibility, and result significance. Based on the evaluation results, the research methods are ranked, and methods that meet the reliability threshold and their corresponding experimental parameters are selected to form a candidate scheme library.
4. The medical automation research method according to claim 3, characterized in that, The process of automatically generating experimental scripts and calling computing resources to execute experiments based on research methods and experimental parameters in the candidate solution library, and obtaining experimental results, includes: The experimental parameters of each research method in the candidate scheme library are analyzed, and the experimental scripts in the corresponding programming languages are generated. The experimental script is executed by calling the computing resource cluster, and training logs and intermediate results are collected in real time during the execution process; After the experiment is completed, the preset evaluation index values are extracted from the output as the experimental results.
5. The medical automation research method according to claim 4, characterized in that, The step of obtaining the corresponding code implementation based on the research methods in the candidate solution library includes: Collect academic papers in the field of artificial intelligence and extract the addresses of publicly available code repositories by parsing the text of the papers. Based on the research methods in the candidate solution library, match the corresponding code repository address, and clone or download the code repository; Parse the directory structure and dependencies of the code repository to generate a code structure description.
6. The medical automation research method according to claim 5, characterized in that, The process of establishing an alignment between research methods and code implementation, and training a document generation model using aligned data, includes: Establish a correspondence between research methods and code implementations, and combine the research method description text and code snippets into training samples; The pre-trained language model is fine-tuned using the training samples, with the training objective being to generate corresponding LaTeX documents based on the research methodology description.
7. The medical automation research method according to claim 6, characterized in that, When the experimental results meet preset conditions, the document generation model is used to integrate the research methods and experimental results to generate an academic document in LaTeX format, including: Determine whether the experimental results exceed the preset performance improvement threshold; If so, the trained document generation model is invoked, and the research method description and experimental results are input to generate a first draft of an academic document in LaTeX format. Adjust the format and verify the content of the first draft, and output the final academic document.
8. A medical automated research device, applied to the medical automated research method according to any one of claims 1 to 7, characterized in that, The device includes: The automatic retrieval module is used to automatically retrieve academic papers related to the target medical research task and build an initial literature database. The screening module is used to extract research methods and their experimental parameters from the papers in the initial literature database, and to screen the research methods based on a preset reliability assessment model to obtain a candidate scheme library. The execution module is used to automatically generate experimental scripts and call computing resources to execute experiments based on the research methods and experimental parameters in the candidate scheme library, and obtain experimental results; The training module is used to obtain the corresponding code implementation based on the research methods in the candidate solution library, establish the alignment relationship between the research methods and the code implementation, and use the alignment data to train the document generation model. An automatic generation module is used to integrate the research methods and experimental results into an academic document in LaTeX format when the experimental results meet preset conditions, using the document generation model.
9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.
10. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.